{"id":"W4416050261","doi":"10.48550/arxiv.2508.14936","title":"Can synthetic data reproduce real-world findings in epidemiology? A replication study using adversarial random forests","year":2025,"lang":"en","type":"preprint","venue":"MDC Repository (Max-Delbrueck-Center for Molecular Medicine)","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Leibniz-Gemeinschaft; Deutsche Forschungsgemeinschaft; European Commission; Bundesministerium für Bildung und Forschung; Freie Hansestadt Bremen","keywords":"Synthetic data; Replication (statistics); Curse of dimensionality; Generative model; Dimensionality reduction; Data quality; Stability (learning theory); Generative grammar; Statistical model","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","open_science"],"consensus_categories":[],"category_scores_codex":[0.01007198,0.001050135,0.002318379,0.001349255,0.0004833416,0.0001509552,0.006083949,0.0006144789,0.00000522273],"category_scores_gemma":[0.008389365,0.001014495,0.0003695489,0.0009700261,0.0002784338,0.0003006828,0.005509274,0.002228966,0.000002198473],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001175055,"about_ca_system_score_gemma":0.001171988,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02326827,"about_ca_topic_score_gemma":0.002201773,"domain_scores_codex":[0.9849153,0.003320451,0.003217525,0.00625734,0.0010221,0.001267294],"domain_scores_gemma":[0.9814901,0.001539704,0.001694426,0.01430668,0.0005677809,0.0004013366],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002073871,0.001877815,0.9270695,0.004226753,0.001275324,0.001632749,0.01237343,0.02215071,0.002392929,0.003741282,0.007215633,0.01396996],"study_design_scores_gemma":[0.01664463,0.001193773,0.0817357,0.008810244,0.001059511,0.000349887,0.0003662273,0.8748308,0.0007946496,0.005759782,0.006069595,0.002385216],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4329904,0.003040089,0.5041405,0.01918834,0.01774771,0.01914741,0.0003012572,0.001128034,0.002316254],"genre_scores_gemma":[0.9739552,0.0000696278,0.02094756,0.001178371,0.001153082,0.0009344012,0.0009796097,0.0001326886,0.0006495356],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8526801,"threshold_uncertainty_score":0.9999634,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09009185104687441,"score_gpt":0.4003850397833321,"score_spread":0.3102931887364577,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}